drtrigon
commented
7 years ago - Enhance file-metadata by integrating head pose estimation and face recognition: https://phabricator.wikimedia.org/T141646
- Enhance file-metadata in core by implementing machine learning methods and computer vision alogrithm training: https://phabricator.wikimedia.org/T141658
It would be nice if we can have a module which uses the detected faces and attempts to recognize the face using a given dataset. This can be used to detect names of atheletes in the upcoming Rio Olympics.
The way this can work is: Get a list of all Olympic athletes available online somewhere. Google search their name with google image search and download the first 100 or 500 images. Now a facial recognition match can be done to find which person it is using this database. (Possibly using opencv's facerec - http://docs.opencv.org/2.4/modules/contrib/doc/facerec/facerec_tutorial.html)
Note that the images from google need not be only portraits as face recognition can find how to crop them to get only face.